Search now ends in an answer as often as a list of links. This course covers both halves of the job — the SEO that still decides whether you are retrievable at all, and the AEO that decides whether you are the source quoted back to the person asking.
⏱ 5 min read
For twenty years the job was simple to describe: rank in a list of ten blue links, earn the click. That list is still there, but it is no longer the first thing most people read, and for a growing share of searches it is not read at all.
| Surface | What it is | How you get in |
|---|---|---|
| Classic results | The ranked list of links | Traditional SEO — relevance, quality, authority |
| AI Overviews | Google's generated summary above the results | Usually drawn from pages already ranking well, with extractable passages |
| AI Mode | Google's conversational search surface | Same retrieval, heavier decomposition into sub-questions |
| Assistants | ChatGPT, Perplexity, Claude, Copilot | Retrieval through a search partner or live fetch, then citation |
The common thread matters more than the differences: every one of these still begins by crawling and indexing ordinary web pages. There is no separate AI index you can submit to. What changes is what happens after retrieval — instead of handing the user a link, the system lifts a passage, paraphrases it, and sometimes names where it came from.
Two effects, pulling in opposite directions. Fewer clicks, because a question answered on the results page does not need a visit — one widely cited clickstream analysis put the share of Google searches ending without any external click at roughly two-thirds. And better clicks, because someone who arrives after reading an answer that named you has already shortlisted you.
For a small business this is less alarming than the headlines suggest, provided you adjust what you measure. Raw sessions on informational blog posts will drift down. Enquiries from people who say "I asked ChatGPT who does this in Cleveland and your name came up" will drift up. If you only watch the first number you will conclude the work stopped paying, while the second number is quietly where your pipeline moved.
Crawlability, indexing, site speed, mobile rendering, and whether anyone credible references you. Every answer engine sits on top of those. A page that Google cannot index is a page no AI system will cite, because the retrieval step never surfaces it. This is why the back half of this course is not a replacement for the front half — the technical modules are the price of admission to the AEO modules.
Three layers, built in order. Retrievable — the page can be crawled, indexed, and rendered without JavaScript gymnastics. Extractable — the answer sits in a passage that survives being cut out of the page. Trusted — a named author, corroborated facts, and a recent date. Skip a layer and the ones above it stop working: beautifully structured content on an unindexed page earns nothing at all.
It helps to be precise about which traffic is at risk. Pages answering a simple factual question — opening hours, a definition, a conversion — lose the most, because the answer fits in the summary and the visit adds nothing. Pages where the visit is the value lose far less: calculators, booking flows, detailed comparisons, anything with a tool or a form at the end of it.
That has a direct planning consequence. Keep publishing the explanatory content, because it is what gets you cited and it is how people first encounter your name — but put your build effort into the pages an answer cannot substitute for. A citation earns the consideration; something worth visiting earns the visit.
⏱ 5 min read
The two disciplines overlap by maybe eighty per cent. Knowing which twenty per cent differs is what stops you paying twice for the same work.
| SEO | AEO | |
|---|---|---|
| Goal | A position in the list | A citation inside the answer |
| Unit of competition | The page | The passage |
| What wins | Relevance, authority, experience | All of that, plus extractability and freshness |
| Success metric | Clicks, impressions, position | Citations, brand mentions, AI referrals |
| Feedback speed | Weeks to months | Structure fixes can show within days; authority still takes months |
| Failure mode | Ranked 14th, nobody scrolls | Ranked 3rd, nobody can lift a clean sentence from you |
Self-contained claims. "A small business website typically costs $3,000 to $15,000" survives being quoted alone. "As we saw above, it depends on those factors" does not. Name the subject in the sentence; do not lean on the paragraph before it.
Answer placement. Put the direct answer in the first screen, then expand. Analyses of cited pages consistently find the quoted passage sits near the top — CXL's review of a hundred AI Overview citations put more than half inside the first third of the source page.
Corroboration. A claim that exists only on your site is an assertion. The same claim echoed by an industry body, a trade publication and a supplier reads as established fact, and systems weighing which source to trust notice the difference.
Running AEO as a separate content programme — a parallel set of "AI pages" alongside your real ones. It splits your authority across two thin sets of pages, doubles your maintenance, and produces exactly the kind of machine-targeted filler that quality systems are built to discount. One content programme, structured so passages stand alone, is the whole strategy.
The two disciplines also differ in how quickly you see movement, which matters when you are deciding what to do first with limited hours.
| Change | Typical time to show |
|---|---|
| Fixing an indexing problem | Days — once recrawled |
| Restructuring a page opening | Days to a couple of weeks — answer engines recrawl often |
| Adding or correcting schema | Days to weeks for rich result eligibility |
| Refreshing stale figures | Weeks — and it decays again, so it is a cadence not a task |
| New content on a new topic | Two to six months |
| Authority and corroboration | Three months and up, compounding |
The practical order falls out of that table: fix what is broken, restructure what already ranks, then build new. Most people do it in exactly the reverse order, which is why six months of publishing sometimes produces nothing — the new pages were added on top of a site that could not be retrieved properly in the first place.
Keep a single content calendar and a single technical backlog. The only thing AEO adds to your process is a second review pass before publishing: read the first screen of the page on its own and ask whether it answers the headline completely. If it does, you are done. That pass takes two minutes and is the whole of the extra work.
⏱ 5 min read
None of this is glamorous and all of it is gating. Work through it once per site and most of it stays fixed for years.
Open Google Search Console and inspect three or four of your most important URLs. The question is binary: is this page indexed? If it is not, the reason is listed, and the common ones are short to fix — a stray noindex tag, a robots.txt rule, a canonical pointing at the wrong page, or content thin enough that Google crawled it and declined. Submit an XML sitemap at the same time; it speeds discovery, though it never forces indexing.
Google executes JavaScript, but rendering is queued and delayed. Many AI crawlers do not execute it at all, or do so less reliably. If your main content only appears after a client-side fetch, an AI crawler may retrieve an empty shell and move on.
The test takes thirty seconds: load your page with JavaScript disabled in the browser, or use Search Console's live test and read the rendered HTML. If the answer you want quoted is not in that text, assume it is invisible. Server-side rendering or static HTML is the safe default — which, conveniently, is how WordPress and Shopify already work.
| Metric | Measures | Target | Usual culprit |
|---|---|---|---|
| LCP | Time until the main content appears | ≤ 2.5s | Oversized hero images, slow hosting |
| INP | Responsiveness to taps and clicks | ≤ 200ms | Too much JavaScript, third-party scripts |
| CLS | Layout shifting as the page loads | ≤ 0.1 | Images without dimensions, late-loading ads and fonts |
Judge these on field data — the real-user measurements in Search Console — not the lab score PageSpeed Insights shows at the top. A site can score 98 in the lab and fail in the field. Page experience is a modest ranking input and a large conversion one, which is the better reason to fix it.
Google indexes the mobile rendering of your site. Content hidden, truncated or broken on a phone is content that may not count. Test on an actual device: tap targets big enough, text readable without zooming, no interstitial covering the answer.
Your robots.txt governs more than Googlebot now. GPTBot and OAI-SearchBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot and Google-Extended all read it. Blocking them is a legitimate choice with an obvious consequence: you will not be cited by what you block. A business trying to be discovered should almost never block them — but you should know what your host or security plugin shipped by default rather than inheriting it blindly.
Three small files do a disproportionate amount of work, and all three are quick to check.
HTTPS is non-negotiable and free. Check that http redirects to https in a single hop, with no chain, and that no asset loads over http. Beyond that, four status codes cover almost everything: 200 for a working page, 301 for a permanent move, 404 for something genuinely gone, and 5xx for a server problem — the only one of the four that is an emergency, because sustained server errors cause Google to crawl you less.
Inspect your five most important URLs in Search Console. Load the homepage with JavaScript disabled. Check the Core Web Vitals report for a failing group. Open robots.txt and read it. Confirm the sitemap loads and is submitted. That pass catches the large majority of what actually blocks a small business site, and it is repeatable quarterly in the same twenty minutes.
⏱ 5 min read
This is the highest-leverage module in the course, and the cheapest to apply to pages you already have.
A short FAQ answering questions the body genuinely did not cover is useful. Twenty FAQ entries bolted onto every page, repeating what is already above them, is padding — and since Google restricted FAQ rich results for most sites, it does not even buy the extra space in the results that it once did. Three to six real questions, each with a real answer, is the right size.
Question headings with no answer beneath them. The same answer published across six near-identical pages. Pages written to be quoted with nothing a human would stay for. All three are detectable, and the third is self-defeating: the citation earns consideration, but the page still has to earn the visit.
You almost never need to start from scratch. The retrofit takes about fifteen minutes per page and follows the same four steps every time.
Do this to your ten highest-value pages before you write anything new. Pages that already rank somewhere on page one or two are where the returns are, because retrieval already favours them — they are simply hard to quote.
Read your opening block and ask what a reader knows after it that they did not know before. If the honest answer is "that this page is about the topic", you have written an introduction rather than an answer. Introductions are a print-era habit; on a page competing for extraction they cost you the one position that matters.
⏱ 5 min read
Structured data is a block of machine-readable code that states plainly what a page contains: this is an article, by this author, published on this date; this is a product, priced at this, in stock; this is a local business, at this address, open these hours.
Humans never see it. It sits in the page source as JSON-LD — the format Google recommends — and it does two jobs. It makes you eligible for rich results: star ratings, prices, breadcrumbs, event dates, the visual extras that make a listing occupy more of the page. And it removes ambiguity for any system trying to work out what you are and what you said.
It does not raise your ranking. Google has said so repeatedly, and the same logic applies to AI citation: schema on a thin page does not make it citable. It is a clarity mechanism, not a lever. The reason to add it is that ambiguity is a reason for a system to pick someone else.
Mark up only what is visible on the page. Describing content a visitor cannot see is a guidelines violation and one of the easier ones for Google to detect. Validate before you ship, with the Rich Results Test. Warnings are optional fields you skipped; errors mean the rich result will not appear at all.
On WordPress, an SEO plugin handles Organization, Article and breadcrumbs with almost no configuration — add FAQ and Service deliberately, on pages that warrant them. On Shopify, Product schema usually ships with the theme; check it outputs the real price. On a custom or Lovable build, write the JSON-LD yourself and validate it. Whatever the platform, resist the plugin setting that sprays markup across every page type; irrelevant schema is noise, and noise is what you were trying to remove.
datePublished and dateModified carry more weight than most people expect, because freshness is a strong citation signal. Many themes bump the modified date whenever any field is touched, producing a site where everything was apparently updated yesterday — a pattern that is easy to spot and worth nothing. Check what your CMS actually outputs.
It helps to know what you are buying. Valid Product markup can put price, availability and review stars in the listing. Breadcrumb markup replaces the raw URL with a readable path. Event, Recipe and Job posting markup unlock dedicated result formats. Organization markup feeds the knowledge panel that sometimes appears for branded searches.
Eligibility is not a guarantee — Google decides per query whether to show a rich result. But the listings that get them occupy more of the screen and attract more clicks, which is a real competitive advantage on a crowded results page, and it costs you a plugin setting rather than a rewrite.
Search Console adds a report per structured data type it detects, showing valid items, warnings and errors. Warnings are recommended fields you skipped and are usually fine to leave; errors mean the rich result will not appear and should be fixed. Check it after any theme or plugin update — markup is one of the first things a redesign quietly breaks, and nobody notices for months because nothing on the page looks different.
⏱ 5 min read
Search and AI systems reason about entities — things with an identity — rather than strings of text. Your company is an entity. So is each of your people, your location, and the services you sell. The work in this module is making sure those entities are unambiguous and consistent everywhere they appear.
It is where a system confirms what kind of organisation you are, where you operate, how long you have existed, and who is behind it. Write it plainly rather than aspirationally: what you do, for whom, where, since when. "CLIKT is a web and content agency in Northeast Ohio building WordPress, Shopify and Lovable sites for small businesses, operating since 1996" is the kind of sentence that gets reused. A page of mission-statement language is not.
NAP is name, address and phone — and the rule is that all three must match exactly everywhere they appear: your site, your Google Business Profile, directory listings, your invoices, your social profiles. "Suite 200" in one place and "Ste. 200" in another is the kind of mismatch that creates doubt about whether two listings describe the same business. This is unglamorous data hygiene and it is one of the strongest local signals there is.
Anonymous content is weaker under every framework now in play. Attach a real person to anything making claims: a name, a bio that establishes why they can speak on the subject, and links to profiles that verify it. For a hands-on business, the credential that matters is usually experience rather than qualifications — "has run this agency since 1996 and built several hundred sites" does more work than any list of certifications.
| Signal | Where it lives | What good looks like |
|---|---|---|
| Organisation identity | About page + Organization schema | One name, stated plainly, with sameAs links |
| Location | Contact page, GBP, footer, LocalBusiness schema | Identical NAP in every instance |
| People | Author bios + Person schema | Real names, real experience, linked profiles |
| Reputation | Reviews, press, directories, partner pages | Independent sources saying the same thing about you |
Google Business Profile reviews influence local ranking and click-through, and they are one of the few places a third party describes your work in public. Ask routinely, respond to all of them including the unflattering ones, and never buy them. For AI answers specifically, reviews matter because they are corroboration: your claim that you are good at something, said by someone who is not you.
Being mentioned matters now even without a link. Industry associations, chamber listings, local press, podcast appearances, supplier and partner pages, conference programmes — each one is another independent source confirming the same entity exists and does what it says. This turns ordinary business development into visibility work, which is why AEO belongs with marketing rather than buried in a technical SEO checklist.
Nine lines, each of which gets reused by something:
⏱ 5 min read
Stop reading and do this one. Twenty minutes with three assistants tells you more about your AI visibility than any tool subscription.
Open ChatGPT, Perplexity and Google AI Mode. Pick five questions a customer would genuinely ask before hiring or buying from you — not your brand name, which proves nothing, but the category questions where you want to be among the named options. Run each question in all three. Record what came back and, crucially, which sources were cited.
Three things to note for each question. Were you named? If not, who was. What got cited? Open those pages and look at where on the page the quoted material sits and what format it took — a definition, a table row, a numbered step. Was anything wrong about you? Outdated services, a wrong location, a stale price. Incorrect information in an answer is usually traceable to incorrect or missing information on your own pages.
Pick the five highest-leverage items from what you found and do them this week:
Answers vary between sessions, accounts and locations, so one run is an anecdote. The same five questions run monthly is a trend line. Keep it in a spreadsheet with the date, the engine, whether you were named, and who else was — that sheet is the only AI visibility baseline most small businesses need, and it costs nothing.
One sheet, five columns: date, engine, question, were you named, who else was. Thirty seconds per row, fifteen rows a month. After three months you can see whether your name appears more often, whether a competitor is pulling ahead, and whether the pages you fixed are now the ones being cited.
Not being named is diagnostic, not just disappointing. If a competitor is cited and you are not, open their page: nine times in ten the difference is structural rather than editorial — they answered the question in the first screen and you buried it, or they put the comparison in a table and you wrote it as prose. That is a fifteen-minute fix on your side, which is why this exercise is worth repeating monthly.
Add a sixth column once you are a month in: what you changed. A log that pairs "restructured the pricing page opening" with "named in three of five answers the following month" is the closest thing to attribution this discipline currently offers.
That sheet is also the single most persuasive thing you can put in front of a client or a boss, because it shows the shift in plain language rather than through a metric that needs explaining. "In January we were named in one of fifteen answers; in April it was six" lands in a way that an impressions chart does not.
⏱ 4 min read
Three layers of measurement, in descending order of reliability.
Clicks, impressions, CTR and average position, straight from Google, with sixteen months of history. The weekly habit: Performance over the last 28 days compared with the previous period, sorted by click difference, plus a glance at the Pages report for new indexing problems. Filter queries to positions 8–20 to find the pages one push away from page one.
Google launched dedicated generative AI performance reports in Search Console in June 2026, rolling out to a subset of properties first. If your property has that view, it is the most trustworthy measurement of AI visibility available to you — first-party impressions and clicks from AI surfaces rather than a vendor's sample.
Search Console tells you about you; Semrush tells you about everyone else. The three uses that justify the subscription for a small agency: keyword gap against two or three competitors, site audit for technical problems at scale, and position tracking for the handful of terms a client actually cares about. Treat volume and difficulty numbers as directional estimates, never as fact — they are models of a dataset only Google holds.
The citation-tracking category is young and expensive, and most of its published research is vendor marketing. The tools sample prompts and report frequency, which is useful for trend direction and unreliable as absolute measurement, because answers vary by user and session. For most small businesses the monthly manual check from module seven, plus AI referral traffic in Analytics, is enough — and the money is better spent on the content.
Lead with outcomes: enquiries from organic, then clicks and impressions, then what changed on the site and when. Keep a change log — it is the only thing that lets you attribute a result three months later. Stop leading with rankings; they vary by person, place and device, and they invite a conversation about a number you cannot control.
Weeks one to four produce indicators, not revenue: indexing problems cleared, pages restructured, maybe an early citation. Weeks four to eight is where restructured pages start showing movement in Search Console and the occasional AI answer starts naming you. Month three is the earliest point at which a trend means anything.
Say that out loud at the start, to yourself or to a client. The work is cheap and reliable; it is only slow. The people who abandon it do so in week six, which is always just before the first signal.
Once the initial fixes are done, this settles into something small and repeatable. Weekly: fifteen minutes in Search Console. Monthly: the five-question assistant check, and one page restructured or refreshed. Quarterly: the twenty-minute technical pass from module three, a schema check after any theme update, and a look at whether your entity details are still consistent everywhere.
⏱ ~6 min
Answer all ten, then score yourself. Explanations appear after scoring.
All consulted October 2026. Search, AI and platform products change fast — check the primary documentation before acting on anything time-sensitive.
You now know what to fix. We do the fixing — technical cleanup, answer-first rewrites, schema and entity signals, measured in Search Console and in the assistants themselves.
Most sites are two weeks of work away from being citable. We set this up in days, not months.
CLIKT.com: Northeast Ohio web, SEO & AI since 1996.